Introduction to Optical Networks

Introduction to Optical Networks

S WE ENTER THE NEW MILLENNIUM, we are seeing dramatic changes in the telecommunications industry that have far-reaching implications for our lifestyles. There are many drivers for these changes. First and foremost is the continuing, relentless need for more capacity in the network. This demand is fueled by many factors. The tremendous growth of the Internet and the World Wide Web, both in terms of number of users as well as the amount of time and thus bandwidth taken by each user, is a major factor. A simple example of the latter phenomenon is the following: An average voice phone call lasts about 3 minutes; in contrast, users connecting to the Internet via dialup lines typically stay on for an average of 20 minutes. So an Internet call brings in about six times as much traffic into a network as a voice call. Internet traffic has been doubling every four to six months, and this trend appears set to continue for a while. Meanwhile we are seeing the ongoing deployment of broadband access technologies such as digital subscriber line (DSL) and cable modems, which provide bandwidths per user on the order of 1 Mb/s, contrasted against the 28-56 kb/s available over dialup lines. The impact of such deployments is quite significant. A 10% increase in penetration of DSL among the 100 million total U.S. households will bring in another 1 Tb/s of traffic into the network, assuming that 10% of these users are on simultaneously. At the same time, businesses today rely on high-speed networks to conduct their businesses. These networks are used to interconnect multiple locations within a company as well as between companies for business-to-business transactions. Large corporations that used to lease 1.5 Mb/s lines to interconnect their internal sites are commonly leasing 155 Mb/s connections today.

A

2

INTRODUCTION TO OPTICAL NETWORKS

There is also a strong correlation between the increase in demand and the cost of bandwidth. Technological advances have succeeded in continously reducing the cost of bandwidth. This reduced cost of bandwidth in turn spurs the development of a new set of applications that make use of more bandwidth and affects behavioral patterns. A simple example is that as phone calls get cheaper, people spend more time on the phone. This in turn drives the need for more bandwidth in the network. This positive feedback cycle shows no sign of abating in the near future. Another factor causing major changes in the industry is the deregulation of the telephone industry and the breaking up of telephone monopolies. For several decades, the telecommunications business was controlled by service providers who were essentially monopolies. In fact this is still the case in many parts of the world. It is a well-known fact that monopolies impede rapid progress. Monopolistic companies can take their time adapting to changes and have no incentive to reduce costs and provide new services. Deregulation of these monopolies has stimulated competition in the marketplace, which in turn has resulted in lower costs to end users and faster deployment of new technologies and services. For example, since the long-distance market was deregulated in the United States in 1984, long-distance phone call rates have dropped by 1.8% annually. In contrast, in the monopoly-dominated local phone market, local phone call rates have increased by 3.6% annually. Deregulation has also resulted in creating a number of new startup service providers as well as startup companies providing equipment to these service providers. This is a big difference from the situation that existed until the mid- to late 1990s, where the telecommunications industry was dominated by a few large service providers and a few large equipment suppliers to these service providers. There is also a significant change in the type of traffic that is increasingly dominating the network. Much of the new demand is being spurred by data, as opposed to traditional voice traffic~a trend that has already existed for quite a while. However, much of the network today is architected to efficiently support voice traffic, not data traffic. This change in traffic mix is causing service providers to reexamine the way they build their networks, the type of services they deliver, and even their entire business model, in many cases. We will study the impact of this later in this chapter and also in Chapter 13. These factors have driven the development of high-capacity optical networks and their remarkably rapid transition from the research laboratories into commercial deployment. This book aims to cover optical network technologies, systems, and networking issues, as well as economic and other deployment considerations.

1.1

1.1

Telecommunications Network Architecture

3

Telecommunications N e t w o r k Architecture Our focus in this book is primarily on the so-called public networks, which are networks operated by service providers, or carriers, as they are often called. Carriers use their network to provide a variety of services to their customers. Carriers used to be essentially telephone companies, but today there are many different breeds of carriers operating under different business models, many of whom do not even provide telephone service. In addition to the traditional carriers providing telephone and leased line services, today there are carriers who are dedicated to interconnecting Internet service providers (ISPs), carriers who are in the business of providing bulk bandwidth to other carriers, and even virtual carriers who provide services without owning any infrastructure. In many cases, the carrier owns the facilities (for example, fiber links) and equipment deployed inside the network. Building fiber links requires right-of-way privileges. Not anybody can dig up streets! Fiber is deployed in many different ways today--buried underground, strung on overhead poles, and buried beside oil and gas pipelines and railroad tracks. In other cases, carriers may lease facilities from other carriers and in turn offer value-added services using these facilities. For example, a long-distance phone service provider may not own a network at all but rather simply buy bandwidth from another carrier and resell it to end users in smaller portions. A local-exchange carrier (LEC) offers local services in metropolitan areas, and an interexchange carrier (IXC) offers long-distance services. This distinction is blurring rapidly as LECs expand into long distance and IXCs expand into local services. In order to understand this better, we need to step back and look at the history of deregulation in the telecommunications services industry. In the United States, before 1984, there was one phone companymAT&T. AT&T, along with the local Bell operating companies, which it owned, held a monopoly for both long-distance and local services. In 1984, with the passing of the telecommunications deregulation act, the overall entity was split into AT&T, which could offer only long-distance services, and a number of "baby" Bells, or regional Bell operating companies (RBOCs), which offered local services and were not allowed to offer long-distance services. Long-distance services were deregulated, and many other companies, such as MCI and Sprint, successfully entered the long-distance market. The baby Bells came to be known as the incumbent LECs (ILECs) and were still monopolies within their local regions. With all the consolidation that has happened in the industry, we are left with four RBOCs, Southwestern Bell Communications (SBC), Bell Atlantic (now Verizon), BellSouth, and U.S. West (now part of Qwest). In addition to the RBOCs,

4

INTRODUCTION TO OPTICAL NETWORKS

there are other competitive LECs (CLECs) that are less regulated and compete with the RBOCs to offer local services. The terminology used above is prevalent mostly in North America. In Europe, we had a similar situation where the government-owned postal, telephone, and telegraph (PTT) companies held monopolies within their respective countries. Over the past decade, deregulation has set in, and we now have a number of new carriers in Europe offering both local and long-distance services. In the rest of the book, we will take a more general approach and classify carriers as metro carriers or long-haul carriers. While the same carrier may offer metro and long-haul services, the networks used to deliver long-haul services are somewhat different from metro networks, and so it is useful to keep this distinction. In contrast to public networks, private networks are networks owned and operated by corporations for their internal use. Many of these corporations in turn rely on capacity provided by public networks to implement their private networks, particularly if these networks cross public land where right-of-way permits are required to construct networks. Networks within buildings spanning at most a few kilometers are called local-area networks (LANs), those that span a campus or metropolitan area, typically tens to a few hundred kilometers, are called metropolitan-area networks (MANs), and networks that span even longer distances, ranging from several hundred to thousands of kilometers, are called wide-area networks (WANs). We will also see a similar type of classification used in public networks, which we study next. Figure 1.1 shows an overview of a typical public fiber network architecture. The network is vast and complex, and different parts of the network may be owned and operated by different carriers. The nodes in the network are central offices, sometimes also called points of presence (POPs). (In some cases, POPs refer to "small" nodes and hubs refer to "large" nodes.) The links between the nodes consist of fiber pairs, and in many cases, multiple fiber pairs. Links in the long-haul network tend to be very expensive to construct. For this reason, the topology of many North American long-haul networks is fairly sparse. In Europe, the link lengths are shorter and the European long-haul network topologies tend to be denser. At the same time, it is imperative to provide alternate paths for traffic in case some of the links fail. These constraints have resulted in the widespread deployment of ring topologies, particularly in North America. Rings are sparse (only two links per node), but still provide an alternate path to reroute traffic. In many cases a meshed network is actually implemented in the form of interconnected ring networks. At a high level, the network can be broken up into a metropolitan (or metro) network and a long-haul network. The metro network is the part of the network that lies within a large city or a region. The long-haul network interconnects cities or different regions. The metro network consists of a metro access network and a

1.1

Telecommunications Network Architecture

5

office

Home

Business

Long haul

Metropolitan P " - -'~L

Interexchange network

Figure 1.1

Metropolitan iv-

Interoffice network

-'~L

Access network

lb. r

Different parts of a public network.

metro interoffice network. The access network extends from a central office out to individual businesses or homes (typically groups of homes rather than individual homes at this time). The access network's reach is typically a few kilometers, and it mostly collects traffic from customer locations into the carrier network. Thus most of the traffic in the access network is hubbed into the carrier's central office. The interoffice network connects groups of central offices within a city or region. This network typically spans a few kilometers to several tens of kilometers between offices. The long-haul network interconnects different cities or regions and spans hundreds to thousands of kilometers between central offices. In some cases, there is another part of the network that provides the handoff between the metro network and the long-haul network, particularly if these networks are operated by different carriers. In contrast to the access network, the traffic distribution in the metro interoffice and long-haul networks is meshed (or distributed). The distances indicated here are illustrative and vary widely based on the location of the network. For example, intercity distances in Europe are often only a few hundred kilometers, whereas intercity distances in North America can be as high as a few thousand kilometers. The network shown in Figure 1.1 is a terrestrial network. Optical fiber is also extensively used in undersea networks. Undersea networks can range from a few hundred kilometers in distance to several thousands of kilometers for routes that cross the Atlantic and Pacific oceans.

6

INTRODUCTION TO OPTICAL NETWORKS

11 II II 'II II----II I

IF__] n2n I-iI[_.__] m2m m-iim__m m2m

mmmmmm (a)

, F-N I

IF--]

I - ~ F--I

oomommomo

(b)

Figure 1.2

1.2

Different types of time division multiplexing: (a) fixed, (b) statistical.

Services, Circuit Switching, and Packet Switching Many types of services are offered by carriers to their customers. In many cases, these are connection-oriented services, in that there is the notion of a connection between two or more parties across an underlying network. The differences lie in the bandwidth of the connection and the type of underlying network with which the connection is supported, which has a significant impact on the quality-of-service guarantees offered by the carriers to their customers. Networks can also provide connectionless service; we will discuss this later in this section. There are two fundamental types of underlying network infrastructures based on how traffic is multiplexed and switched inside the network: circuit-switched and packet-switched. Figure 1.2 illustrates some of the differences in the type of multiplexing used in these cases. A circuit-switched network provides circuit-switched connections to its customers. In circuit switching, a guaranteed amount of bandwidth is allocated to each connection and available to the connection all the time, once the connection is set up. The sum of the bandwidth of all the circuits, or connections, on a link must be less than the link bandwidth. The most common example of a circuit-switched network is the public-switched telephone network (PSTN), which provides a nailed-down

1.2

Services, Circuit Switching, and Packet Switching

7

connection to end users with a fixed amount of bandwidth (typically around 4 kHz) once the connection is established. This circuit is converted to a digital 64 kb/s circuit at the carrier central office. This network was designed to support voice streams and does a fine job for this application. The circuit-switched services offered by carriers today include circuits at a variety of bit rates, ranging from 64 kb/s voice circuits all the way up to several Gb/s. These connections are typically leased by a carrier to its customers and remain nailed down for fairly long periods, ranging from several days to months to years as the bandwidth on the connection goes up. These services are also called private line services. The PSTN fits into this category with one important difference~in the PSTN, users dial up and establish connections between themselves, whereas with private line services, the carrier usually sets up the connection using a management system. This situation is changing, and we will no doubt see users dialing for higher-speed private lines in the future, particularly as the connection durations come down. The problem with circuit switching is that it is not efficient at handling bursty data traffic. An example of a bursty traffic stream is traffic from a user typing on a keyboard. When the user is actively typing, bits are transmitted at more or less a steady rate. When the user pauses, there is no traffic. Another example is Web browsing. When a user is looking at a recently downloaded screen, there is almost no traffic. When she clicks on a hyperlink, a new page needs to be downloaded as soon as possible from the network. Thus a bursty stream requires a lot of bandwidth from the network whenever it is active and very little bandwidth when it is not active. It is usually characterized by an average bandwidth and a peak bandwidth, which correspond to the long-term average and the short-term burst rates, respectively. In a circuit-switched network, we would have to reserve sufficient bandwidth to deal with the peak rate, and this bandwidth would be unused a lot of the time. Packet switching was invented to deal with the problem of tranporting bursty data traffic efficiently. In packet-switched networks, the data stream is broken up into small packets of data. These packets are multiplexed together with packets from other data streams inside the network. The packets are switched inside the network based on their destination. To facilitate this switching, a packet header is added to the payload in each packet. The header carries addressing information, for example, the destination address or the address of the next node in the path. The intermediate nodes read the header and determine where to switch the packet based on the information contained in the header. At the destination, packets belonging to a particular stream are received and the data stream is put back together. The predominant example of a packet-switched network is the Internet, which uses the Internet Protocol (IP) to route packets from their source to their destination. Packet switching uses a technique called statistical multiplexing when multiplexing multiple bursty data streams together on a link. Since each data stream is bursty,

8

INTRODUCTION TO OPTICAL NETWORKS

it is likely that at any given time only some streams are active and others aren't. The probability that all streams are active simultaneously is quite small. Therefore the bandwidth required on the link can be made significantly smaller than the bandwidth that would be required if all streams were to be active simultaneously. Statistical multiplexing improves the bandwidth utilization but leads to some other important effects. If more streams are active simultaneously than there is bandwidth available on the link, some packets will have to be queued or buffered until the link becomes free again. The delay experienced by a packet therefore depends on how many packets are queued up ahead of it. This causes the delay to be a random parameter. On occasion, the traffic may be so high that it causes the buffers to overflow. When this happens, some of the packets must be dropped from the network. Usually, a higher-layer transport protocol, such as the transmission control protocol (TCP) in the Internet, detects this and ensures that these packets are retransmitted. On top of this, a traditional packet-switched network does not even support the notion of a connection. Packets belonging to a connection are treated as independent entities, and different packets may take different routes through the network. This is the case with networks using IP. This type of connectionless service is called a datagram service. This leads to even more variations in the delays experienced by different packets and also forces the higher-layer transport protocol to resequence packets that arrive out of sequence at their destinations. Thus, traditionally, such a packet-switched network provides what is called besteffort service. The network tries its best to get data from its source to its destination as quickly as possible, but offers no guarantees. This is indeed the case with much of the Internet today. Another example of this type of service is frame relay. Frame relay is a popular packet-switched service provided by carriers to interconnect corporate data networks. When a user signs up for frame relay service, she is promised a certain average bandwidth over time but allowed to have an instantaneous burst rate above this rate, however without any guarantees. In order to ensure that the network is not overloaded, the user data rate may be regulated at the input to the network so that the user does not exceed her committed average bandwidth over time. In other words, a user who is provided a committed rate of 64 kb/s may send data at 128 kb/s on occasion, and 32 kb/s at other times, but will not be allowed to exceed the average rate of 64 kb/s over a long period of time. This best-effort service provided by packet-switched networks is fine for a number of applications, such as Web browsing and file transfers, which are not highly delay-sensitive applications. However, applications such as real-time video or voice calls cannot tolerate random packet delays. Therefore, there is a great deal of effort today to design packet-switched networks that can provide some guarantees on the quality of service that they offer. Examples of quality of service (QoS) may include certain guarantees on the maximum packet delay as well as the variation in the delay,

1.2

Services, Circuit Switching, and Packet Switching

and guarantees on providing a minimum average bandwidth for each connection. The asynchronous transfer mode (ATM) network is a consequence of this thinking. The Internet Protocol has also been enhanced to provide similar services. Most of these QoS efforts rely on the notion of having a connection-oriented layer. For example, in an IP network, multi-protocol label switching (MPLS) provides virtual circuits to support end-to-end traffic streams. A virtual circuit forces all packets belonging to that circuit to follow the same path through the network, allowing better allocation of resources in the network to meet certain quality-of-service guarantees, such as bounded delay for each packet. Unlike a real circuit-switched network, a virtual circuit does not provide a fixed guaranteed bandwidth along the path of the circuit due to the fact that statistical multiplexing is used to multiplex virtual circuits inside the network.

1.2.1

The Changing Services Landscape The service model used by the carriers is changing rapidly as networks and technologies evolve and competition among carriers intensifies. The bandwidth delivered per connection is increasing, and it is becoming common to lease lines ranging in capacity from 155 Mb/s to 2.5 Gb/s and even 10 Gb/s. Note that in many cases, a carrier's customer is another carrier. The so-called carrier's carrier essentially delivers bandwidth in large quantities to interconnect other carriers' networks. Also, because of increased competition and customer demands, carriers now need to be able to deliver these connections rapidly in minutes to hours rather than days to months, once the bandwidth is requested. Moreover, rather than signing up for contracts that range from months to years, customers would like to sign up for much shorter durations. It is not unthinkable to have a situation where a user leases a large amount of bandwidth for a relatively short period of time, for example, to perform large backups at certain times of the day, or to handle special events, or to handle temporary surges in demands. Another aspect of change has to do with the availability of these circuits, which is defined as the percentage of time the service is available to the user. Typically, carriers provide 99.999% availability, which corresponds to a downtime of less than five minutes per year. This in turn requires the network to be designed to provide very fast restoration of service in the event of failures such as fiber cuts, today in about 50 ms. While this will remain true for a subset of connections, there will be other connections carrying data that may be able to tolerate higher restoration times. There may be connections that may not need to be restored at all by the carrier, with the user dealing with rerouting traffic on these connections in the event of failures. Very fast restoration is usually accomplished by providing full redundancymhalf the bandwidth in the network is reserved for this purpose. We will see in Chapter 10

10

INTrODUCTiON TO OPTICAL NETWORI~S

that more sophisticated techniques can be used to improve the bandwidth efficiency but usually at the cost of slower restoration times. This realization is stimulating the development of service offerings that trade off restoration time against bandwidth efficiency in the network. Thus carriers in the new world need to deploy networks that provide them with the flexibility to deliver bandwidth on demand when needed where needed, with the appropriate service attributes. The "where needed" is significant because carriers can rarely predict the location of future traffic demands. As a result it is difficult for them to plan and build networks optimized around specific assumptions on bandwidth demands. At the same time, the mix of services offered by carriers is expanding. We talked about different circuit-switched and packet-switched services earlier. What is not commonly realized is that today these services are delivered over separate overlay networks, rather than a single network. Thus carriers need to operate and maintain multiple networksua very expensive proposition over time. For most networks, the costs associated with operating the network over time (such as maintenance, provisioning of new connections, upgrades) far outweigh the initial cost of putting in the equipment to build the network. Carriers would thus like to migrate to maintaining a single network infrastructure that enables them to deliver multiple types of services.

1.3

Optical Networks Optical networks offer the promise to solve many of the problems discussed above. In addition to providing enormous capacities in the network, an optical network provides a common infrastructure over which a variety of services can be delivered. These networks are also increasingly becoming capable of delivering bandwidth in a flexible manner where and when needed. Optical fiber offers much higher bandwidth than copper cables and is less susceptible to various kinds of electromagnetic interferences and other undesirable effects. As a result, it is the preferred medium for transmission of data at anything more than a few tens of megabits per second over any distance more than a kilometer. It is also the preferred means of realizing short-distance (a few meters to hundreds of meters), high-speed (gigabits per second and above) interconnections inside large systems. The latest statistics from the U. S. Federal Communications Commission [Kra99] indicate the ubiquity of fiber deployment. Optical fibers are widely deployed today in all kinds of telecommunications networks, except perhaps in residential access networks. Although fiber is provided to many businesses today, particularly in large cities, it has yet to reach individual homes, due to the huge cost of wiring the infrastructure and the questionable rate of return on this investment seen by the

1.3

11

Optical Networks

/x

1,000,000 10,000 Long haul 100 Leased lines

Local-area networks

o

0.01 t Residential access

0.0001 !

.

1980

1985

.

. 1990

. 1995

2000

Year

Figure 1.3

Bandwidth growth over time in different types of networks.

service providers. Before providing some more data, we introduce some terminology first. Each route in a network comprises many fiber cables. Each cable contains many fibers. For example, a 10-mile-long route using 3 fiber cables is said to have 10 route miles and 30 sheath (cable) miles. If each cable has 20 fibers, the same route is said to have 600 fiber miles. As of the end of 1998, the local-exchange carriers in the United States had deployed more than 355,000 sheath miles of fiber, containing more than 16 million fiber miles. More than 160,000 route miles of fiber had been deployed by the interexchange carriers in the United States, containing more than 3.6 million miles of optical fiber [Kra99]. Fiber transmission technology has evolved over the past few decades to offer higher and higher bit rates on a fiber over longer and longer distances. Figure 1.3 plots the growth in bandwidth over time of different types of networks, updated from a chart originally presented in [Fra93]. This tremendous growth in bandwidth is primarily due to the deployment of optical fiber communication systems. Note from Figure 1.3 that the bandwidths into our homes are still limited by the bandwidth available on our phone line, which is made of twisted-pair copper cable. These lines are capable of carrying data at a few megabits per second using digital subscriber loop (DSL) technology, but voice-grade lines are limited at the central office to 4 kHz of bandwidth. (Note that in some cases we measure bandwidth in hertz and sometimes loosely use bit rates when talking about bandwidth. We will

12

INTRODUCTION TO OPTICAL NETWORKS

quantify the relationship between bit rate and bandwidth in Section 1.7.) The other alternative is the cable network, which again is capable of providing a few megabits per second to each subscriber on a shared basis using cable modem technology. When we talk about optical networks, we are really talking about two generations of optical networks. In the first generation, optics was essentially used for transmission and simply to provide capacity. Optical fiber provided lower bit error rates and higher capacities than copper cables. All the switching and other intelligent network functions were handled by electronics. Examples of first-generation optical networks are SONET (synchronous optical network) and the essentially similar SDH (synchronous digital hierarchy) networks, which form the core of the telecommunications infrastructure in North America and in Europe and Asia, respectively, as well as a variety of enterprise networks such as ESCON (enterprise serial connection). We will study these first-generation networks in Chapter 6. Today we are seeing the deployment of second-generation optical networks, where some of the routing, switching, and intelligence is moving into the optical layer. Before we discuss this new generation of networks, we will first look at the multiplexing techniques that provide the capacity needed to realize these networks.

1.3.1

Multiplexing Techniques The need for multiplexing is driven by the fact that it is much more economical to transmit data at higher rates over a single fiber than it is to transmit at lower rates over multiple fibers, in most applications. There are fundamentally two ways of increasing the transmission capacity on a fiber, as shown in Figure 1.4. The first is to increase the bit rate. This requires higher-speed electronics. Many lower-speed data streams are multiplexed into a higher-speed stream at the transmission bit rate by means of electronic time division multiplexing (TDM). The multiplexer typically interleaves the lower-speed streams to obtain the higher-speed stream. For example, it could pick 1 byte of data from the first stream, the next byte from the second stream, and so on. As an example, 64 155 Mb/s streams may be multiplexed into a single 10 Gb/s stream. Today, the highest transmission rate in commercially available systems is around 10 Gb/s; 40 Gb/s TDM technology will be available soon. To push TDM technology beyond these rates, researchers are working on methods to perform the multiplexing and demultiplexing functions optically. This approach is called optical time division multiplexing (OTDM). Laboratory experiments have demonstrated the multiplexing/demultiplexing of several 10 Gb/s streams into/from a 250 Gb/s stream, although commercial implementation of OTDM is still several years away. We will study OTDM systems in Chapter 12. However, multiplexing and demultiplexing high-speed streams by itself is not sufficient to realize practical networks. We need to contend with the various impairments that arise as these very

1.3

Optical Networks

13

Figure 1.4 Different multiplexing techniques for increasing the transmission capacity on an optical fiber. (a) Electronic or optical time division multiplexing and (b) wavelength division multiplexing. Both multiplexing techniques take in N data streams, each of B b/s, and multiplex them into a single fiber with a total aggregate rate of NB b/s.

high-speed streams are transmitted over a fiber. As we will see in Chapters 5 and 13, the higher the bit rate, the more difficult it is to engineer around these impairments. However, similar bottlenecks have been encountered in the past, and people have always found ways to overcome them; so we can expect the transmission bit rates to continue to increase, although perhaps not at the breakneck pace of the past two decades. Another way to increase the capacity is by a technique called wavelength division multiplexing (WDM). W D M is essentially the same as frequency division multiplexing (FDM), which has been used in radio systems for more than a century. For some reason, the term FDM is used widely in radio communication, but W D M is used in

14

INTRODUCTION TO OPTICAL NETWORKS

the context of optical communication, perhaps because FDM was studied first by communications engineers and WDM by physicists. The idea is to transmit data simultaneously at multiple carrier wavelengths (or, equivalently, frequencies or colors) over a fiber. To first order, these wavelengths do not interfere with each other provided they are kept sufficiently far apart. (There are some undesirable second-order effects where wavelengths do interfere with each other, and we will study these in Chapters 2 and 5.) Thus WDM provides virtual fibers, in that it makes a single fiber look like multiple "virtual" fibers, with each virtual fiber carrying a single data stream. WDM systems are widely deployed today in long-haul and undersea networks and are being deployed in metro networks as well. WDM and TDM both provide ways to increase the transmission capacity and are complementary to each other. Therefore networks today use a combination of TDM and WDM. The question of what combination of TDM and WDM to use in systems is an important one facing carriers today. For example, suppose a carrier wants to install an 80 Gb/s link. Should we deploy 32 WDM channels at 2.5 Gb/s each or should we deploy 10 WDM channels at 8 Gb/s each? The answer depends on a number of factors, including the type and parameters of the fiber used in the link and the services that the carrier wishes to provide using that link. We will discuss this issue in Chapter 13. Using a combination of WDM and TDM, systems with transmission capacities of around I Tb/s over a single fiber are becoming commercially available, and no doubt we will see systems with higher capacities operating over longer distances emerge in the future.

1.3.2

Second-Generation Optical Networks Optics is clearly the preferred means of transmission, and WDM transmission is now widely used in the network. In recent years, people have realized that optical networks are capable of providing more functions than just point-to-point transmission. Major advantages are to be gained by incorporating some of the switching and routing functions that were performed by electronics into the optical part of the network. For example, as data rates get higher and higher, it becomes more difficult for electronics to process data. Suppose the electronics must process data in blocks of 53 bytes each (this happens to be the block size in asynchronous transfer mode networks). In a 100 Mb/s data stream, we have 4.24 lZS to process a block, whereas at 10 Gb/s, the same block must be processed within 42.4 ns. In first-generation networks, the electronics at a node must handle not only all the data intended for that node but also all the data that is being passed through that node on to other nodes in the network. If the latter data could be routed through in the optical domain, the burden on the underlying electronics at the node would be significantly reduced. This is one of the key drivers for second-generation optical networks.

1.3

Optical Networks

15

Optical networks based on this paradigm are now being deployed. The architecture of such a network is shown in Figure 1.5. We call this network a wavelength-routing network. The network provides lightpaths to its users, such as SONET terminals or IP routers. Lightpaths are optical connections carried end to end from a source node to a destination node over a wavelength on each intermediate link. At intermediate nodes in the network, the lightpaths are routed and switched from one link to another link. In some cases, lightpaths may be converted from one wavelength to another wavelength as well along their route. Different lightpaths in a wavelength routing network can use the same wavelength as long as they do not share any common links. This allows the same wavelength to be reused spatially in different parts of the network. For example, Figure 1.5 shows six lightpaths. The lightpath between B and C, the lightpath between D and E, and one of the lightpaths between E and F do not share any links in the network and can therefore be set up using the same wavelength )~1. At the same time, the lightpath between A and F shares a link with the lightpath between B and C and must therefore use a different wavelength. Likewise, the two lightpaths between E and F must be assigned different wavelengths. Note that these lightpaths all use the same wavelength on every link in their path. This is a constraint that we must deal with if we do not have wavelength conversion capabilities within the network. Suppose we had only two wavelengths available in the network and wanted to set up a new lightpath between nodes E and E Without wavelength conversion, we would not be able to set up this lightpath. On the other hand, if the intermediate node X can perform wavelength conversion, then we can set up this lightpath using wavelength )~2 on link EX and wavelength ~1 on link XE The key network elements that enable optical networking are optical line terminals (OLTs), optical add~drop multiplexers (OADMs), and optical crossconnects (OXCs), as shown in Figure 1.5. An OLT multiplexes multiple wavelengths into a single fiber and demultiplexes a set of wavelengths on a single fiber into separate fibers. OLTs are used at the ends of a point-to-point WDM link. An OADM takes in signals at multiple wavelengths and selectively drops some of these wavelengths locally while letting others pass through. It also selectively adds wavelengths to the composite outbound signal. An OADM has two line ports where the composite WDM signals are present, and a number of local ports where individual wavelengths are dropped and added. An OXC essentially performs a similar function but at much larger sizes. OXCs have a large number of ports (ranging from a few tens to thousands) and are able to switch wavelengths from one input port to another. Both OADMs and OXCs may incorporate wavelength conversion capabilities. The detailed architecture of these networks will be discussed in Chapter 7. Optical networks based on the architecture described above are already being deployed. OLTs have been widely deployed for point-to-point applications. OADMs

16

INTRODUCTION TO OPTICAL NETWORKS

Figure 1.5 A WDM wavelength-routing network, showing optical line terminals (OLTs), optical add/drop multiplexers (OADMs), and optical crossconnects (OXCs). The network provides lightpaths to its users, which are typically IP routers or SONET terminals.

are now used in long-haul and metro networks. OXCs are beginning to be deployed first in long-haul networks because of the higher capacities in those networks.

1.4

The Optical Layer Before delving into the details of the optical layer, we first introduce the notion of a layered network architecture. Networks are complicated entities with a variety of different functions being performed by different components of the network, with equipment from different vendors all interworking together. In order to simplify our view of the network, it is desirable to break up the functions of the network into different layers, as shown in Figure 1.6. This type of layered model was proposed by the International Standards Organization (ISO) in the early 1980s. Imagine the layers as being vertically stacked up. Each layer performs a certain set of functions

1.4

The Optical Layer

17

Figure 1.6 Layered hierarchy of a network showing the layers at each network element (NE).

and provides a certain set of services to the next higher layer. In turn, each layer expects the layer below it to deliver a certain set of services to it. The service interface between two adjacent layers is called a service access point (SAP), and there can be multiple SAPs between layers corresponding to different types of services offered. In most cases, the network provides connections to the user. A connection is established between a source and a destination node. Setting up, taking down, and managing the state of a connection is the job of a separate network control and management entity (not shown in Figure 1.6), which may control each individual layer in the network. There are also examples where the network provides connectionless services to the user. These services are suitable for transmitting short messages across a network, without having to pay the overhead of setting up and taking down a connection for this purpose. We will confine the following discussion to the connection-oriented model. Within a network element, data belonging to a connection flows between the layers. Each layer multiplexes a number of higher-layer connections and may add some additional overhead to data coming from the higher layer. Each intermediate network element along the path of a connection embodies a set of layers starting from the lowest layer up to a certain layer in the hierarchy.

18

INTRODUCTION TO OPTICAL NETWORKS

Figure 1.7 The classical layered hierarchy.

It is imi0ortant to define the functions of each layer and the interfaces between layers. This is essential because it allows vendors to manufacture a variety of hardware and software products performing the functions of some, but not all, of the layers, and provide the appropriate interfaces to communicate with other products performing the functions of other layers. There are many possible implementations and standards for each layer. A given layer may work together with a variety of lower or higher layers. Each of the different types of optical networks that we will study constitutes a layer. Each layer itself can in turn be broken up into several sublayers. As we study these networks, we will explore this layered hierarchy further. Figure 1.7 shows a classical breakdown of the different layers in a network that was proposed by the International Standards Organization. The lowest layer in the hierarchy is the physical layer, which provides a "pipe" with a certain amount of bandwidth to the layer above it. The physical layer may be optical, wireless, or coaxial or twisted-pair cable. The next layer above is the data link layer, which is responsible for framing, multiplexing, and demultiplexing data sent over the physical layer. The framing protocol defines how data is transported over a physical link. Typically data is broken up into frames before being transmitted ove~ a physical link. This is necessary to ensure reliable delivery of data across the link. The framing protocol provides clear delineation between frames, provides sufficient transitions in the signal so that it can be recovered at the other end, and usually includes additional overhead that enables link errors to be detected. Examples of data link protocols suitable for operation over point-to-point links include the point-to-point protocol (PPP) and the high-level data link control (HDLC) protocol. Included in the data link layer is the

1.4

The Optical Layer

19

media access control layer (MAC), which coordinates the transmissions of different nodes when they all share common bandwidth, as is the case in many local-area networks, such as Ethernets and token rings. Above the data link layer resides the network layer. The network layer usually provides virtual circuits or datagram services to the higher layer. A virtual circuit (VC) represents an end-to-end connection with a certain set of quality-of-service parameters associated with it, such as bandwidth and error rate. Data transmitted by the source over a VC is delivered in sequence at its destination. Datagrams, on the other hand, are short messages transmitted end to end, with no notion of a connection. The network layer performs the end-to-end routing function of taking a message at its source and delivering it to its destination. The predominant network layer today is IP, and the main network element in an IP network is an IP router. IP provides a way to route packets (or datagrams) end to end in a packet-switched network. IP includes statistical multiplexing of multiple packet streams and today also provides some simple and relatively slow and inefficient service restoration mechanisms. The Internet Protocol has been adapted to operate over a variety of data link and physical media, such as Ethernet, serial telephone lines, coaxial cable lines, and optical fiber lines. The transport layer resides on top of the network layer and is responsible for ensuring the end-to-end, in-sequence, and error-free delivery of the transmitted messages. For example, the transmission control protocol (TCP) used in the Internet belongs to this layer. Above the transport layer reside other layers such as the session, presentation, and application layers, but we will not be concerned with these layers in this book. Another important packet-switched layer is ATM. ATM provides a connectionoriented service (virtual circuits) and is capable of providing a variety of quality-of-service guarantees. Packets in ATM are called cells and are of fixed length (53 bytes). ATM is being used by many carriers as a vehicle to deliver reliable packet-switched services. More on this subject in Chapters 6 and 13. This classical layered view of networks needs some embellishment to handle the variety of networks and protocols that are proliferating today. A more realistic layered model for today's networks would employ multiple protocol stacks residing one on top of the other. Each stack incorporates several sublayers, which may provide functions resembling traditional physical, link, and network layers. To provide a concrete example of this, consider an IP over SONET network shown in Figure 1.8. In this case, the IP network treats the SONET network as providing it with point-to-point links between IP routers. The SONET layer itself, however, internally routes and switches connections, and in a sense, incorporates its own link, physical, and network layers.

20

INTRODUCTION TO OPTICAL NETWORKS

Figure 1.8 An IP over SONET network. (a) The network has IP switches with SONET adaptors that are connected to a SONET network. (b) The layered view of this network.

Figure 1.9

The layered view of an IP over ATM over SONET network.

Another example of this sort of layering arises in the context of an IP over ATM over SONET network. Some service providers are deploying an ATM network operating over a SONET infrastructure to provide services for IP users. In such a network, IP packets are converted to ATM cells at the periphery of the network. The ATM switches are connected through a SONET infrastructure. The layered view of such a network is shown in Figure 1.9. Again, the IP network treats the ATM network as its link layer, and the ATM network uses SONET as its link layer. The introduction of second-generation optical networks adds yet another layer to the protocol hierarchy--the so-called optical layer. The optical layer is a s e r v e r layer

1.4

The Optical Layer

21

Figure 1.10 A layered view of a network consisting of a second-generation optical network layer that supports a variety of client layers above it.

that provides services to other client layers. This optical layer provides lightpaths to a variety of client layers, as shown in Figure 1.10. Examples of client layers residing above a second-generation optical network layer include IP, ATM, and SONET/SDH, as well as other possible protocols such as Gigabit Ethernet, ESCON (enterprise serial connection~a protocol used to interconnect computers to storage devices and other computers), or Fibre Channel (which performs the same function as ESCON, at higher speeds). As second-generation optical networks evolve, they may provide other services besides lightpaths, such as packet-switched virtual circuit or datagram services. These services may directly interface with user applications, as shown in Figure 1.10. Several other layer combinations are possible and not shown in the figure, such as IP over SONET over optical, and ATM over optical. The client layers make use of the lightpaths provided by the optical layer. To a SONET network operating over the optical layer, the lightpaths are simply replacements for hardwired fiber connections between SONET terminals. As described earlier, a lightpath is a connection between two nodes in the network, and it is set up by assigning a dedicated wavelength to it on each link in its path. Note that individual wavelengths are likely to carry data at fairly high bit rates (a few gigabits per second), and this entire bandwidth is provided to the higher layer by a lightpath. Depending on the capabilities of the network, this lightpath could be set up or taken down in response to a request from the higher layer. This can be thought of as a circuit-switched service, akin to the service provided by today's telephone network: the network sets up or takes down calls in response to a request from the user. Alternatively, the network may provide only permanent lightpaths, which are set up

22

INTRODUCTION TO OPTICAL NETWORKS

at the time the network is deployed. This lightpath service can be used to support high-speed connections for a variety of overlying networks. Optical networks today provide functions that might be thought of as falling primarily within the physical layer from the perspective of its users. However, the optical network itself incorporates several sublayers, which in turn correspond to the link and network layer functions in the classical layered view. Before the emergence of the optical layer, SONET/SDH was the predominant transmission layer in the telecommunications network, and it is still the dominant layer in many parts of the network. We will study SONET/SDH in detail in Chapter 6. For convenience, we will use SONET terminology in the rest of this section. The SONET layer provides several key functions. It provides end-to-end, managed, circuit-switched connections. It provides an efficient mechanism for multiplexing lower-speed connections into higher-speed connections. For example, low-speed voice connections at 64 kb/s or private line 1.5 Mb/s connections can be multiplexed all the way up into 2.5 Gb/s or 10 Gb/s line rates for transport over the network. Moreover, at intermediate nodes, SONET provides an efficient way to extract individual low-speed streams from a high-speed stream, using an elegant multiplexing mechanism based on the use of pointers. SONET also provides a high degree of network reliability and availability. Carriers expect their networks to provide 99.99% to 99.999% of availability. These numbers translate into an allowable network downtime of less than one hour per year and five minutes per year, respectively. SONET achieves this by incorporating sophisticated mechanisms for rapid service restoration in the event of failures in the network. This is a subject we will look at in Chapter 10. Finally, SONET includes extensive overheads that allow operators to monitor and manage the network. Examples of these overheads include parity check bytes to determine whether frames are received in error or not, and connection identifiers that allow connections to be traced and verified across a complex network. SONET network elements include line terminals, add/drop multiplexers (ADMs), regenerators, and digital crossconnects (DCSs). Line terminals multiplex and demultiplex traffic streams. ADMs are deployed in linear and ring network configurations. They provide an efficient way to drop part of the traffic at a node while allowing the remaining traffic to pass through. The ring topology allows traffic to be rerouted around failures in the network. Regenerators regenerate the SONET signal wherever needed. DCSs are deployed in larger nodes to switch a large number of traffic streams. Today's DCSs are capable of switching thousands of 45 Mb/s traffic streams. The functions performed by the optical layer are in many ways analogous to those performed by the SONET layer. The optical layer multiplexes multiple lightpaths into a single fiber and allows individual lightpaths to be extracted efficiently from the composite multiplex signal at network nodes. It incorporates sophisticated service

1.4

The Optical Layer

23

Figure 1.11 Example of a typical multiplexing layered hierarchy.

restoration techniques and management techniques as well. We will look at these techniques in Chapters 9 and 10. Figure 1.11 shows a typical layered network hierarchy, highlighting the optical layer. The optical layer provides lightpaths that are used by SONET and IP network elements. The SONET layer multiplexes low-speed circuit-switched streams into higher-speed streams, which are then carried over lightpaths. The IP layer performs statistical multiplexing of packet-switched streams into higher-speed streams, which are also carried over lightpaths. Inside the optical layer itself is a multiplexing hierarchy. Multiple wavelengths or lightpaths are combined together into wavelength bands. Bands are combined together to produce a composite WDM signal on a fiber. The network itself may include multiple fibers and multiple fiber bundles, each of which carries a number of fibers. So why have multiple layers in the network that perform similar functions? The answer is that this form of layering significantly reduces network equipment costs. Different layers are more efficient at performing functions at different bit rates. For example, the SONET layer can efficiently (that is, cost-effectively) switch and process traffic streams up to, say, 2.5 Gb/s today. However, it is very expensive to have this layer process 100 10 Gb/s streams coming in on a WDM link. The optical layer, on the other hand, is particularly efficient at processing traffic on a wavelength-by-wavelength basis, but not particularly good at processing traffic streams at lower granularities, for example, 155 Mb/s. Therefore, it makes sense to use the optical layer to process large amounts of bandwidth at a relatively coarse level and the SONET layer to process smaller amounts of bandwidth at a relatively finer

24

INTRODUCTION TO OPTICAL NETWORKS

level. This fundamental observation is the key driver to providing such functions in multiple layers, and we will study this in detail in Chapter 7. A similar observation also holds for the service restoration function of these networks. Certain failures are better handled by the optical layer and certain others by the SONET layer or the IP layer. We will study this aspect in Chapter 10.

1.5

Transparency and All-Optical Networks A major feature of the lightpath service provided by second-generation networks is that this type of service can be transparent to the actual data being sent over the lightpath once it is set up. For instance, a certain maximum and minimum bit rate might be specified, and the service may accept data at any bit rate and any protocol format within these limits. It may also be able to carry analog data. Transparency in the network provides several advantages. An operator can provide a variety of different services using a single infrastructure. We can think of this as service transparency. Second, the infrastructure is future-proof in that if protocols or bit rates change, the equipment deployed in the network is still likely to be able to support the new protocols and/or bit rates without requiring a complete overhaul of the entire network. This allows new services to be deployed efficiently and rapidly, while allowing legacy services to be carried as well. An example of a transparent network of this sort is the telephone network. Once a call is established in the telephone network, it provides 4 kHz of bandwidth over which a user can send a variety of different types of traffic such as voice, data, or fax. There is no question that transparency in the telephone network today has had a far-reaching impact on our lifestyles. Transparency has become a useful feature of second-generation optical networks as well. Another term associated with transparent networks is the notion of an all-optical network. In an all-optical network, data is carried from its source to its destination in optical form, without undergoing any optical-to-electrical conversions along the way. In an ideal world, such a network would be fully transparent. However, all-optical networks are limited in their scope by several parameters of the physical layer, such as bandwidth and signal-to-noise ratios. For example, analog signals require much higher signal-to-noise ratios than digital signals. The actual requirements depend on the modulation format used as well as the bit rate. We will study these aspects in Chapter 5, where we will see that engineering the physical layer is a complex task with a variety of parameters to be taken into consideration. For this reason, it is very difficult to build and operate a network that can support analog as well as digital signals at arbitrary bit rates.

1.5 Transparency and All-Optical Networks

25

The other extreme is to build a network that handles essentially a single bit rate and protocol (say, 2.5 Gb/s SONET only). This would be a nontransparent network. In between is a practical network that handles digital signals at a range of bit rates up to a specified maximum. Most optical networks being deployed today fall into this category. Although we talk about optical networks, they almost always include a fair amount of electronics. First, electronics plays a crucial role in performing the intelligent control and management functions within a network. However, even in the data path, in most cases, electronics is needed at the periphery of the network to adapt the signals entering the optical network. In many cases, the signal may not be able to remain in optical form all the Way to its destination due to limitations imposed by the physical layer design and may have to be regenerated in between. In other cases, the signal may have to be converted from one wavelength to another wavelength. In all these situations, the signal is usually converted from optical form to electronic form and back again to optical form' Having these electronic regenerators in the path of the signal reduces the transparency of that path. There are three types of electronic regeneration techniques for digital data. The standard one is called regeneration with retiming and reshaping, also known as 3R. Here the bit clock is extracted from the signal, and the signal is reclocked. This technique essentially produces a "fresh" copy of the signal at each regeneration step, allowing the signal to go through a very large number of regenerators. However, it eliminates transparency to bit rates and the framing protocols, since acquiring theclock usually requires knowledge of both of these. Some limited form of bit rate transparency is possible by making use of programmable clock recovery chips that can work at a set of bit rates that are multiples of one another. For example, chipsets that perform clock recovery at either 2.5 Gb/s or 622 Mb/s are commercially available today. An implementation using regeneration of the optical signal without retiming, also called 2R, offers transparency to bit rates, without supporting analog data or different modulation formats [GJR96]. However, this approach limits the number of regeneration steps allowed, particularly at higher bit rates, over a few hundred megabits per second. The limitation is due to the jitter, which accumulates at each regeneration step. The final form of electronic regeneration is 1R, where the signal is simply received and retransmitted without retiming or reshaping. This form of regeneration can handle analog data as well, but its performance is significantly poorer than the other two forms of regeneration. For this reason, the networks being deployed today use 2R or 3R electronic regeneration. Note, however, that optical amplifiers are widely used to amplify the signal in the optical domain, without converting the signal to the electrical domain. These can be thought of as 1R optical regenerators.

26

INTRODUCTION TO OPTICAL NETWORKS

Table 1.1 Different types of transparency in an optical network. Transparency type Parameter

Fully transparent

Practical

Nontransparent

Analog/digital Bit rate Framing protocol

Both Arbitrary Arbitrary

Digital Predetermined maximum Selected few

Digital Fixed Single

Table 1.1 provides an overview of the different dimensions of transparency. At one end of the spectrum is a network that operates at a fixed bit rate and framing protocol, for example, SONET at 2.5 Gb/s. This would be truly an opaque network. In contrast, a fully transparent network would support analog and digital signals with arbitrary bit rates and framing protocols. As we argued earlier, however, such a network is not practical to engineer and build. Today, a practical alternative is to engineer the network to support a variety of digital signals up to a predetermined maximum bit rate and a specific set of framing protocols, such as SONET and Gigabit Ethernet. The network supports a variety of framing protocols either by making use of 2R regeneration inside the network or by providing specific 3R adaptation devices for each of the framing protocols. Such a network is shown in Figure 1.12. It can be viewed as consisting of islands of all-optical subnetworks with optical-to-electrical-to-optical conversion at their boundaries for the purposes of adaptation, regeneration, or wavelength conversion.

1.6

Optical Packet Switching So far we have talked about optical networks that provide lightpaths. These networks are essentially circuit-switched. Researchers are also working on optical networks that can perform packet switching in the optical domain. Such a network would be able to offer virtual circuit services or datagram services, much like what is provided by ATM and IP networks. With a virtual circuit connection, the network offers what looks like a circuit-switched connection between two nodes. However, the bandwidth offered on the connection can be smaller than the full bandwidth available on a link or wavelength. For instance, individual connections in a future high-speed network may operate at 10 Gb/s, while transmission bit rates on a wavelength could be 100 Gb/s. Thus the network must incorporate some form of time division multiplexing to combine multiple connections onto the transmission bit rate. At these rates, it may be easier to do the multiplexing in the optical domain rather than in the electronic domain. This form of optical time domain multiplexing (OTDM) may

1.6

Optical Packet Switching

27

Figure 1.12 An optical network consisting of all-optical subnetworks interconnected by optical-to-electrical-to-optical (OEO) converters. OEO converters are used in the network for adapting external signals to the optical network, for regeneration, and for wavelength conversion.

be fixed or statistical. Those that perform statistical multiplexing are called optical packet-switched networks. For simplicity we will talk mostly about optical packet switching. Fixed OTDM can be thought of as a subset of optical packet switching where the multiplexing is fixed instead of statistical. An optical packet-switching node is shown in Figure 1.13. The idea is to create packet-switching nodes with much higher capacities than can be envisioned with electronic packet switching. Such a node takes a packet coming in, reads its header, and switches it to the appropriate output port. The node may also impose a new header on the packet. It must also handle contention for output ports. If two packets coming in on different ports need to go out on the same output port, one of the packets must be buffered, or sent out on another port. Ideally, all the functions inside the node would be performed in the optical domain, but in practice, certain functions, such as processing the header and controlling the switch, get relegated to the electronic domain. This is because of the very limited processing capabilities in the optical domain. The header itself could be sent at a lower bit rate than the data so that it can be processed electronically. The mission of optical packet switching is to enable packet-switching capabilities at rates that cannot be contemplated using electronic packet switching. However, designers are handicapped by several limitations with respect to processing signals in the optical domain. One important factor is the lack of optical random access

28

INTRODUCTION TO OPTICAL NETWORKS

Figure 1.13 An optical packet-switching node. The node buffers the incoming packets, looks at the packet header, and routes the packets to an appropriate output port based on the information contained in the header.

memory for buffering. Optical buffers are realized by using a length of fiber and are just simple delay lines, not fully functional memories. Packet switches include a high amount of intelligent real-time software and dedicated hardware to control the network and provide quality-of-service guarantees, and these functions are ,difficult to perform in the optical domain. Another factor is the relatively primitive state of fast optical-switching technology, compared to electronics. For these reasons, optical packet switching is still in its infancy today in research laboratories. Chapter 12 covers all these aspects in detail.

1.7

Transmission Basics In this section, we introduce and define the units for common parameters associated with optical communication systems.

1.7.1

Wavelengths, Frequencies, and Channel Spacing When we talk about WDM signals, we will be talking about the wavelength, or frequency, of these signals. The wavelength )~ and frequency f are related by the equation

c=f~., where c denotes the speed of light in free space, which is 3 x 108 m/s. We will reference all parameters to free space. The speed of light in fiber is actually somewhat lower (closer to 2 x 108 m/s), and the wavelengths are also correspondingly different.

1.7

Transmission Basics

29

To characterize a WDM signal, we can use either its frequency or wavelength interchangeably. Wavelength is measured in units of nanometers (nm) or micrometers (#m or microns). (1 nm = 10-9 m, 1 #m = 10 -6 m.) The wavelengths of interest to optical fiber communication are centered around 0.8, 1.3, and 1.55 ~m. These wavelengths lie in the infrared band, which is not visible to the human eye. Frequencies are measured in units of hertz (or cycles per second), more typically in megahertz (1 MHz = 106 Hz), gigahertz (1 GHz = 109 Hz), or terahertz (1 THz = 1012 Hz). Using c - 3 • 108 m/s, a wavelength of 1.55 #m would correspond to a frequency of approximately 193 THz, which is 193 • 1012 Hz. Another parameter of interest is the channel spacing, which is the spacing between two wavelengths or frequencies in a WDM system. Again the channel spacing can be measured in units of wavelengths or frequencies. The relationship between the two can be obtained starting from the equation C

f

~

--o

Differentiating this equation around a center wavelength )~0, we obtain the relationship between the frequency spacing A f and the wavelength spacing A)~ as r

Af -- -- )--~A)~. This relationship is accurate as long as the wavelength (or frequency) spacing is small compared to the actual channel wavelength (or frequency), which is usually the case in optical communication systems. At a wavelength )~0 = 1550 nm, a wavelength spacing of 0.8 nm corresponds to a frequency spacing of 100 GHz, a typical spacing in WDM systems. Digital information signals in the time domain can be viewed as a periodic sequence of pulses, which are on or off, depending on whether the data is a 1 or a 0. The bit rate is simply the inverse of this period. These signals have an equivalent representation in the frequency domain, where the energy of the signal is spread across a set of frequencies. This representation is called the power spectrum, or simply spectrum. The signal bandwidth is a measure of the width of the spectrum of the signal. The bandwidth can also be measured either in the frequency domain or in the wavelength domain, but is mostly measured in units of frequency. Note that we have been using the term bandwidth rather loosely. The bandwidth and bit rate of a digital signal are related but not exactly the same. Bandwidth is usually specified in kilohertz or megahertz or gigahertz, whereas bit rate is specified in kilobits/second (kb/s), megabits/second (Mb/s), or gigabits/second (Gb/s). The relationship between the two depends on the type of modulation used. For instance, a phone line offers 4 kHz of bandwidth, but sophisticated modulation technology allows us to realize

30

INTRODUCTION TO OPTICAL NETWORKS

k "

00 GH IT M

1 bandwidth

00 GH Vl-~

r

193.3

193.2

193.1

193.0

192.9

Frequency (THz)

1550.918

1551.721

1552.524

1553.329

1554.134

Wavelength(nm)

Figure 1.14 The 100 GHz ITU frequency grid based on a reference frequency of 193.1 THz. A 50 GHz grid has also been defined around the same reference frequency.

a bit rate of 56 kb/s over this phone line. This ratio of bit rate to available bandwidth is called spectral efficiency. Optical communication systems use rather simple modulation techniques that achieve a spectral efficiency of about 0.4 bits/s/Hz, and it is reasonable to assume therefore that a signal at a bit rate of 10 Gb/s uses up bandwidth of approximately 25 GHz. Note that the signal bandwidth needs to be sufficiently smaller than the channel spacing; otherwise we would have undesirable interference between adjacent channels and distortion of the signal itself.

1.7.2

Wavelength Standards WDM systems today primarily use the 1.55 >m wavelength region for two reasons: the inherent loss in optical fiber is the lowest in that region, and excellent optical amplifiers are available in that region. We will discuss this in more detail in later chapters. The wavelengths and frequencies used in WDM systems have been standardized on a frequency grid by the International Telecommunications Union (ITU). It is an infinite grid centered at 193.1 THz, a segment of which is shown in Figure 1.14. The ITU decided to standardize the grid in the frequency domain based on equal channel spacings of 50 GHz or 100 GHz. Observe that if multiple channels are spaced apart equally in wavelength, they are not spaced apart exactly equally in frequency, and vice versa. The figure also shows the power spectrum of two channels 400 GHz apart in the grid populated by traffic-bearing signals, as indicated by the increased signal bandwidth on those channels. The ITU grid only tells part of the story. Today, we are already starting to see systems using 25 GHz channel spacings. We are also seeing the use of several transmission bands. The early WDM systems used the so-called C-band, or conventional band (approximately 1530-1565 nm). The use of the L-band, or long wavelength

1.7

Transmission Basics

31

band (approximately 1565-1625 nm), has become feasible recently with the development of optical amplifiers in this band. We will look at this and other bands in Section 1.8. It has proven difficult to obtain agreement from the different WDM vendors and service providers on more concrete wavelength standards. As we will see in Chapters 2 and 5, designing WDM transmission systems is a complex endeavor, requiring trade-offs among many different parameters, including the specific wavelengths used in the system. Different WDM vendors use different methods for optimizing their system designs, and converging on a wavelength plan becomes difficult as a result. However, the ITU grid standard has helped accelerate the deployment of WDM systems because component vendors can build wavelength-selective parts to a specific grid, which helps significantly in inventory management and manufacturing.

1.7.3

Optical Power and Loss In optical communication, it is quite common to use decibel units (dB) to measure power and signal levels, as opposed to conventional units. The reason for doing this is that powers vary over several orders of magnitude in a system, and this makes it easier to deal with a logarithmic rather than a linear scale. Moreover, by using such a scale, calculations that involve multiplication in the conventional domain become additive operations in the decibel domain. Decibel units are used to represent relative as well as absolute values. To understand this system, let us consider an optical fiber link. Suppose we transmit a light signal with power Pt watts (W). In terms of dB units, we have

( Pt )dBW

=

10 log(Pt)w.

In many cases, it is more convenient to measure powers in milliwatts (mW), and we have an equivalent dBm value given as

( Pt )dBm -- 10 log(Pt )mW. For example, a power of 1 mW corresponds to 0 dBm or - 3 0 dBW. A power of 10 mW corresponds to 10 dBm or - 2 0 dBW. As the light signal propagates through the fiber, it is attenuated; that is, its power is decreased. At the end of the link, suppose the received power is P~. The link loss y is then defined as P~ y

---

n o

Pt

32

INTRODUCTION TO OPTICAL NETWORKS

In dB units, we would have (Y)dB = 10 log g = (Pr)dBm -- (Pt)dBm-

Note that dB is used to indicate relative values, whereas dBm and dBW are used to indicate the absolute power value. As an example, if Pt = 1 mW and Pr = 1 /~W, implying that y = 0.001, we would have, equivalently,

(et)dBm =

0

(er)dBm =

- 3 0 dBm or - 60 dBW,

dBm or - 30 dBW,

and (Y)dB = - - 3 0 dB.

In this context, a signal being attenuated by a factor of 1000 would equivalently undergo a 30 dB loss. A signal being amplified by a factor of 1000 would equivalently have a 30 dB gain. We measure loss in optical fiber usually in units of dB/km. So, for example, a light signal traveling through 120 km of fiber with a loss of 0.25 dB/km would be attenuated by 30 dB.

1.8

N e t w o r k Evolution We conclude this chapter by outlining the trends and factors that have shaped the evolution of optical fiber transmission systems and networks. Figure 1.15 gives an overview. The history of optical fiber transmission has been all about how to transmit data at the highest capacity over the longest possible distance and is remarkable for its rapid progress. What is equally remarkable is the fact that researchers have successfully overcome numerous obstacles along this path, many of which when first discovered looked as though they would impede further increases in capacity and transmission distance. The net result of this is that capacity continues to grow in the network, while the cost per bit transmitted per kilometer continues to get lower and lower, to a point where it has become practical for carriers to price circuits independently of the distance. We will introduce various types of fiber propagation impairments as well as optical components in this section. These will be covered in depth in Chapters 2, 3, and 5.

1.8

Network Evolution

33

Figure 1.15 Evolution of optical fiber transmission systems. (a) An early system using LEDs over multimode fiber. (b) A system using MLM lasers over single-mode fiber in the 1.3/~m band to overcome intermodal dispersion in multimode fiber. (c) A later system using the 1.55/~m band for lower loss, and using SLM lasers to overcome chromatic dispersion limits. (d) A current-generation WDM system using multiple wavelengths at 1.55/2m and optical amplifiers instead of regenerators. The P-k curves to the left of the transmitters indicate the power spectrum of the signal transmitted.

1.8.1

Early Days--Multimode Fiber Early experiments in the mid-1960s demonstrated that information encoded in light signals could be transmitted over a glass fiber waveguide. A waveguide provides a medium that can guide the light signal, enabling it to stay focused for a reasonable distance without being scattered. This allows the signal to be received at the other

34

INTRODUCTION TO OPTICAL NETWORKS

end with sufficient strength so that the information can be decoded. These early experiments proved that optical transmission over fiber was feasible. An optical fiber is a very thin cylindrical glass waveguide consisting of two parts: an inner core material and an outer cladding material. The core and cladding are designed so as to keep the light signals guided inside the fiber, allowing the light signal to be transmitted for reasonably long distances before the signal degrades in quality. It was not until the invention of low-loss optical fiber in the early 1970s that optical fiber transmission systems really took off. This silica-based optical fiber has three low-loss windows in the 0.8, 1.3, and 1.55/~m infrared wavelength bands. The lowest loss is around 0.25 dB/km in the 1.55/~m band, and about 0.5 dB/km in the 1.3/~m band. These fibers enabled transmission of light signals over distances of several tens of kilometers before they needed to be regenerated. A regenerator converts the light signal into an electrical signal and retransmits a fresh copy of the data as a new light signal. The early fibers were the so-called multimode fibers. Multimode fibers have core diameters of about 50 to 85/2m. This diameter is large compared to the operating wavelength of the light signal. A basic understanding of light propagation in these fibers can be obtained using the so-called geometrical optics model, illustrated in Figure 1.16. In this model, a light ray bounces back and forth in the core, being reflected at the core-cladding interface. The signal consists of multiple light rays, each of which potentially takes a different path through the fiber. Each of these different paths corresponds to a propagation mode. The length of the different paths is different, as seen in the figure. Each mode therefore travels with a slightly different speed compared to the other modes. The other key devices needed for optical fiber transmission are light sources and receivers. Compact semiconductor lasers and light-emitting diodes (LEDs) provided practical light sources. These lasers and LEDs were simply turned on and off rapidly to transmit digital (binary) data. Semiconductor photodetectors enabled the conversion of the light signal back into the electrical domain. The early telecommunication systems (late 1970s through the early 1980s) used multimode fibers along with LEDs or laser transmitters in the 0.8 and 1.3/~m wavelength bands. LEDs were relatively low-power devices that emitted light over a fairly wide spectrum of several nanometers to tens of nanometers. A laser provided higher output power than an LED and therefore allowed transmission over greater distances before regeneration. The early lasers were multilongitudinal mode (MLM) Fabry-Perot lasers. These MLM lasers emit light over a fairly wide spectrum of several nanometers to tens of nanometers. The actual spectrum consists of multiple spectral lines, which can be thought of as different longitudinal modes, hence the

1.8

Network Evolution

35

Figure 1.16 Geometrical optics model to illustrate the propagation of light in an optical fiber. (a) Cross section of an optical fiber. The fiber has an inner core and an outer cladding, with the core having a slightly higher refractive index than the cladding. (b) Longitudinal view. Light rays within the core hitting the core-cladding boundary are reflected back into the core by total internal reflection.

term MLM. Note that these longitudinal laser modes are different from the propagation modes inside the optical fiber! While both LEDs and MLM lasers emit light over a broad spectrum, the spectrum of an LED is continuous, whereas the spectrum of an MLM laser consists of many periodic lines. These early systems had to have regenerators every few kilometers to regenerate the signal. Regenerators were expensive devices and continue to be expensive today, so it is highly desirable to maximize the distance between regenerators. In this case, the distance limitation was primarily due to a phenomenon known as intermodal dispersion. As we saw earlier, in a multimode fiber, the energy in a pulse travels in different modes, each with a different speed. At the end of the fiber, the different modes arrive at slightly different times, resulting in a smearing of the pulse. This smearing in general is called dispersion, and this specific form is called intermodal dispersion. Typically, these early systems operated at bit rates ranging from 32 to 140 Mb/s with regenerators every 10 km. Such systems are still used for low-cost computer interconnection at a few hundred megabits per second over a few kilometers.

1.8.2

Single-Mode Fiber The next generation of systems deployed starting around 1984 used single-mode fiber as a means of eliminating intermodal dispersion, along with MLM Fabry-Perot lasers in the 1.3 #m wavelength band. Single-mode fiber has a relatively small core diameter of about 8 to 10 #m, which is a small multiple of the operating wavelength

36

INTRODUCTION TO OPTICAL NETWORKS

range of the light signal. This forces all the energy in a light signal to travel in the form of a single mode. Using single-mode fiber effectively eliminated intermodal dispersion and enabled a dramatic increase in the bit rates and distances possible between regenerators. These systems typically had regenerator spacings of about 40 km and operated at bit rates of a few hundred megabits per second. At this point, the distance between regenerators was limited primarily by the fiber loss. The next step in this evolution in the late 1980s was to deploy systems in the 1.55 t~m wavelength window to take advantage of the lower loss in this window, relative to the 1.3/zm window. This enabled longer spans between regenerators. At this point, another impairment, namely, chromatic dispersion, started becoming a limiting factor as far as increasing the bit rates was concerned. Chromatic dispersion is another form of dispersion in optical fiber (we looked at intermodal dispersion earlier). As we saw in Section 1.7, the energy in a light signal or pulse has a finite bandwidth. Even in a single-mode fiber, the different frequency components of a pulse propagate with different speeds. This is due to the fundamental physical properties of the glass. This effect again causes a smearing of the pulse at the output, just as with intermodal dispersion. The wider the spectrum of the pulse, the more the smearing due to chromatic dispersion. The chromatic dispersion in an optical fiber depends on the wavelength of the signal. It turns out that without any special effort, the standard silica-based optical fiber has essentially no chromatic dispersion in the 1.3/~m band, but has significant dispersion in the 1.55/zm band. Thus chromatic dispersion was not an issue in the earlier systems at 1.3/zm. The high chromatic dispersion at 1.55/zm motivated the development of dispersion-shifted fiber. Dispersion-shifted fiber is carefully designed to have zero dispersion in the 1.55/zm wavelength window so that we need not worry about chromatic dispersion in this window. However, by this time there was already a large installed base of standard single-mode fiber deployed for which this solution could not be applied. Some carriers, particularly NTT in Japan and MCI (now part of Worldcom) in the United States, did deploy dispersion-shifted fiber. At this time, researchers started looking for ways to overcome chromatic dispersion while still continuing to make use of standard fiber. The main technique that came into play was to reduce the width of the spectrum of the transmitted pulse. As we saw earlier, the wider the spectrum of the transmitted pulse, the greater the smearing due to chromatic dispersion. The bandwidth of the transmitted pulse is at least equal to its modulation bandwidth. On top of this, however, the bandwidth may be determined entirely by the width of the spectrum of the transmitter used. The MLM Fabry-Perot lasers, as we said earlier, emitted over a fairly wide spectrum of several nanometers (or, equivalently, hundreds of gigahertz), which is much larger than the modulation bandwidth of the signal itself. If we reduce the spectrum of the transmitted pulse to something close to its modulation bandwidth, the penalty due

1.8

Network Evolution

37

to chromatic dispersion is significantly reduced. This motivated the development of a laser source with a narrow spectral widthmthe distributed-feedback (DFB) laser. A DFB laser is an example of a single-longitudinal mode (SLM) laser. An SLM laser emits a narrow single-wavelength signal in a single spectral line, in contrast to MLM lasers whose spectrum consists of many spectral lines. This technological breakthrough spurred further increases in the bit rate to more than 1 Gb/s.

1.8.3

Optical Amplifiers and WDM The next major milestone in the evolution of optical fiber transmission systems was the development of erbium-doped fiber amplifiers (EDFAs) in the late 1980s and early 1990s. The EDFA basically consists of a length of optical fiber, typically a few meters to tens of meters, doped with the rare earth element erbium. The erbium atoms in the fiber are pumped from their ground state to an excited state at a higher energy level using a pump source. An incoming signal photon triggers these atoms to come down to their ground state. In the process, each atom emits a photon. Thus incoming signal photons trigger the emission of additional photons, resulting in optical amplification. Due to a unique coincidence of nature, the difference in energy levels of the atomic states of erbium line up with the 1.5 #m low-loss window in the optical fiber. The pumping itself is done using a pump laser at a lower wavelength than the signal because photons with a lower wavelength have higher energies and energy can be transferred only from a photon of higher energy to that with a lower energy. The EDFA concept was invented in the 1960s but had to wait for the availability of reliable high-power semiconductor pump lasers in the late 1980s and early 1990s before becoming commercially viable. EDFAs spurred the deployment of a completely new generation of systems. A major advantage of EDFAs is that they are capable of amplifying signals at many wavelengths simultaneously. This provided another way of increasing the system capacity: rather than increasing the bit rate, keep the bit rate the same and use more than one wavelength; that is, use wavelength division multiplexing. EDFAs were perhaps the single biggest catalyst aiding the deployment of WDM systems. The use of WDM and EDFAs dramatically brought down the cost of long-haul transmission systems and increased their capacity. At each regenerator location, a single optical amplifier could replace an entire array of expensive regenerators, one per fiber. This proved to be so compelling that almost every long-haul carrier has widely deployed amplified WDM systems today. Moreover WDM provided the ability to turn on capacity quickly, as opposed to the months to years it could take to deploy new fiber. WDM systems with EDFAs were deployed starting in the mid-1990s and are today achieving capacities over 1 Tb/s over a single fiber. At the same time, transmission bit rates on a single channel have risen to 10 Gb/s. Among the earliest

38

INTRODUCTION TO OPTICAL NETWORKS

WDM systems deployed were AT&T's 4-wavelength long-haul system in 1995 and IBM's 20-wavelength MuxMaster metropolitan system in 1994. With the advent of EDFAs, chromatic dispersion again reared its ugly head. Instead of regenerating the signal every 40 to 80 km, signals were now transmitted over much longer distances because of EDFAs, leading to significantly higher pulse smearing due to chromatic dispersion. Again, researchers found several techniques to deal with chromatic dispersion. The transmitted spectrum could be reduced further by using an external device to turn the laser on and off (called external modulation), instead of directly turning the laser on and off (called direct modulation). Using external modulators along with DFB lasers and EDFAs allowed systems to achieve distances of about 600 km at 2.5 Gb/s between regenerators over standard single-mode fiber at 1.55/~m. This number is substantially less at 10 Gb/s. The next logical invention was to develop chromatic dispersion compensation techniques. A variety of chromatic dispersion compensators were developed to compensate for the dispersion introduced by the fiber, allowing the overall residual dispersion to be reduced to within manageable limits. These techniques have enabled commercial systems to achieve distances of several thousand kilometers between regenerators at bit rates as high as 10 Gb/s per channel. At the same time, several other impairments that were second- or third-order effects earlier began to emerge as first-order effects. Today, this list includes nonlinear effects in fiber, the nonflat gain spectrum of EDFAs, and various polarization-related effects. There are several types of nonlinear effects that occur in optical fiber. One of them is called four-wave mixing (FWM). In FWM, three light signals at different wavelengths interact in the fiber to create a fourth light signal at a wavelength that may overlap with one of the light signals. As we can imagine, this signal interferes with the actual data that is being transmitted on that wavelength. It turns out paradoxically that the higher the chromatic dispersion, the lower the effect of fiber nonlinearities. Chromatic dispersion causes the light signals at different wavelengths to propagate at different speeds in the fiber. This in turn causes less overlap between these signals, as the signals go in and out of phase with each other, reducing the effect of the FWM nonlinearity. The realization of this trade-off between chromatic dispersion and fiber nonlinearities stimulated the development of a variety of new types of single-mode fibers to manage the interaction between these two effects. These fibers are tailored to provide less chromatic dispersion than conventional fiber but, at the same time, reduce nonlinearities. We devote Chapter 5 to the study of these impairments and how they can be overcome; we discuss the origin of many of these effects in Chapter 2. Today we are seeing the development of high-capacity amplified terabits/second WDM systems with hundreds of channels at 10 Gb/s, with channel spacings as low as 50 GHz, with distances between electrical regenerators extending to a few thousand

1.8

39

Network Evolution

Table 1.2 Different wavelength bands in optical fiber. The ranges are approximate and have not yet been standardized. Band

Descriptor

Wavelength range (nm)

O-band E-band S-band C-band L-band U-band

Original Extended Short Conventional Long Ultra-long

1260 1360 1460 1530 1565 1625

to to to to to to

1360 1460 1530 1565 1625 1675

kilometers. Systems operating at 40 Gb/s channel rates are in the research laboratories, and no doubt we will see them become commercially available soon. Meanwhile, recent experiments have achieved terabit/second capacities and stretched the distance between regenerators to several thousand kilometers [Cai01, Bak01, VPM01], or achieved total capacities of over 10 Tb/s [Fuk01, Big01] over shorter distances. Table 1.2 shows the different bands available for transmission in single-mode optical fiber. The early WDM systems used the C-band, primarily because that was where EDFAs existed. Today we have EDFAs that work in the L-band, which allow WDM systems to use both the C- and L-bands. We are also seeing the use of other types of amplification (such as Raman amplification, a topic that we will cover in Chapter 3) that complement EDFAs and hold the promise to open up other fiber bands such as the S-band and the U-band for WDM applications. Meanwhile, the development of new fiber types is also opening up a new window in the so-called E-band. This band was previously not feasible due to the high fiber loss in this wavelength range. New fibers have now been developed that reduce the loss in this range. However, there are still no good amplifiers in this band, so the E-band is useful mostly for short-distance applications.

1.8.4

Beyond Transmission Links to Networks The late 1980s also witnessed the emergence of a variety of first-generation optical networks. In the data communications world, we saw the deployment of metropolitan-area networks, such as the 100 Mb/s fiber distributed data interface (FDDI), and networks to interconnect mainframe computers, such as the 200 Mb/s enterprise serial connection (ESCON). Today we are seeing the proliferation of storage networks using the 1 Gb/s Fibre Channel standard for similar applications. In the telecommunications world, standardization and mass deployment of SONET in North America and the similar SDH network in Europe and Japan began. All these

40

INTRODUCTION TO OPTICAL NETWORKS

networks are now widely deployed. Today it is common to have high-speed optical interfaces on a variety of other devices such as IP routers and ATM switches. As these first-generation networks were being deployed in the late 1980s and early 1990s, people started thinking about innovative network architectures that would use fiber for more than just transmission. Most of the early experimental efforts were focused on optical networks for local-area network applications, but the high cost of the technology for these applications has hindered commercial viability of such networks. Research activity on optical packet-switched networks and local-area optical networks continues today. Meanwhile, wavelength-routing networks became a major focus area for several researchers in the early 1990s as people realized the benefits of having an optical layer. Optical add/drop multiplexers and crossconnects are now available as commercial products and are beginning to be introduced into telecommunications networks, stimulated by the fact that switching and routing high-capacity connections is much more economical at the optical layer than in the electrical layer. At the same time, the optical layer is evolving to provide additional functionality, including the ability to set up and take down lightpaths across the network in a dynamic fashion, and the ability to reroute lightpaths rapidly in case of a failure in the network. A combination of these factors is resulting in the introduction of intelligent optical ring and mesh networks, which provide lightpaths on demand and incorporate built-in restoration capabilities to deal with network failures. There was also a major effort to promote the concept of fiber to the home (FTTH) and its many variants, such as fiber to the curb (FTTC), in the late 1980s and early 1990s. The problems with this concept were the high infrastructure cost and the questionable return on investment resulting from customers' reluctance to pay for a bevy of new services such as video to the home. However, telecommunications deregulation, coupled with the increasing demand for broadband services such as Internet access and video on demand, is accelerating the deployment of such networks by the major operators today. Both telecommunications carriers and cable operators are deploying fiber deeper into the access network and closer to the end user. Large businesses requiring very high capacities are being served by fiber-based SONET/SDH or Ethernet networks, while passive optical networks are emerging as possible candidates to provide high-speed services to homes and small businesses. This is the subject of Chapter 11.

Summary We started this chapter by describing the changing face of the telecom industry--the large increase in traffic demands, the increase in data traffic relative to voice traffic, the deregulation of the telecom industry, the resulting emergence of a new set of

Further Reading

41

carriers as well as equipment suppliers to these carriers, the need for new and flexible types of services, and an infrastructure to support all of these. We described two generations of optical networks in this chapter: first-generation networks and second-generation networks. First-generation networks use optical fiber as a replacement for copper cable to get higher capacities. Second-generation networks provide circuit-switched lightpaths by routing and switching wavelengths inside the network. The key elements that enable this are optical line terminals (OLTs), optical add/drop multiplexers (OADMs), and optical crossconnects (OXCs). Optical packet switching may develop over time but faces several technological hurdles. We saw that there were two complementary approaches to increasing transmission capacity: using more wavelengths on the fiber (WDM) and increasing the bit rate (TDM). We also traced the historical evolution of optical fiber transmission and networking. What is significant is that we are still far away from hitting the fundamental limits of capacity in optical fiber. While there are several roadblocks along the way, we will no doubt see the invention of new techniques that enable progressively higher and higher capacities, and the deployment of optical networks with increasing functionality.

Further Reading The communications revolution is a topic that is receiving a lot of coverage across the board these days from the business press. A number of journal and magazine special issues have been focused on optical networks [GLM+00, CSH00, DYJ00, DL00, Alf99, HSS98, CHK+96, FGO+96, HD97, Bar96, NO94, KLHN93, CNW90, Pru89, Bra89]. Several conferences cover optical networks. The main ones are the Optical Fiber Communication Conference (OFC), Supercomm, and the National Fiber-Optic Engineers' Conference. Other conferences such as Next-Generation Networks (NGN), Networld-Interop, European Conference on Optical Communication (ECOC), IEEE Infocom, and the IEEE's International Conference on Communication (ICC) also cover optical networks. Archival journals such as the IEEE's Journal of Lightwave

Technology, Journal of Selected Areas in Communication, Journal of Quantum Electronics, Journal of Selected Topics in Quantum Electronics, Transactions on Networking, and Photonics Technology Letters, and magazines such as the IEEE Communications Magazine and Optical Networks Magazine provide good coverage of this subject. There are several excellent books devoted to fiber optic transmission and components, ranging from fairly basic [Hec98, ST91] to more advanced [KK97a, KK97b,

42

INTRODUCTION TO OPTICAL NETWORKS

Agr97, Agr95, MK88, Lin89]. The 1993 book by Green [Gre93] provides specific coverage of WDM components, transmission, and networking aspects. The historical evolution of transmission systems described here is also covered in a few other places in more detail. [Hec99] is an easily readable book devoted to the early history of fiber optics. [Wil00] is a special issue consisting of papers by many of the optical pioneers providing overviews and historical perspectives of various aspects of lasers, fiber optics, and other component and transmission technologies. [AKW00, Gla00, BKLW00] provide excellent, although Bell Labs-centric, overviews of the historical evolution of optical fiber technology and systems leading up to the current generation of WDM technology and systems. See also [MK88, Lin89]. Kao and Hockham [KH66] were the first to propose using low-loss glass fiber for optical communication. The processes used to fabricate low-loss fiber today were first reported in [KKM70] and refined in [Mac74]. [Sta83, CS83, MT83, Ish83] describe some of the early terrestrial optical fiber transmission systems. [RT84] describes one of the early undersea optical fiber transmission systems. See also [KM98] for a more recent overview. Experiments reporting more than 1 Tb/s transmission over a single fiber were first reported at the Optical Fiber Communication Conference in 1996, and the numbers are being improved upon constantly. See, for example, [CT98, Ona96, Gna96, Mor96, Yan96]. Recent work on these frontiers has focused on (1) transmitting terabits-per-second aggregate traffic across transoceanic distances with individual channel data rates at 10 or 20 Gb/s [Cai01, Bak01, VPM01], or 40 Gb/s channel rates over shorter distances [Zhu01], or (2) obtaining over 10 Tb/s transmission capacity using 40 Gb/s channel rates over a few hundred kilometers [Fuk01, Big01]. Finally, we didn't cover standards in this chaptermbut we will do so in Chapters 6, 9, and 10. The various standards bodies working on optical networking include the International Telecommunications Union (ITU), the American National Standards Institute (ANSI), the Optical Internetworking Forum (OIF), Internet Engineering Task Force (IETF), and Telcordia Technologies. Appendix C provides a list of relevant standards documents.

References [Agr95] G.P. Agrawal. Nonlinear Fiber Optics, 2nd edition. Academic Press, San Diego, CA, 1995. [Agr97] G.P. Agrawal. Fiber-Optic Communication Systems. John Wiley, New York, 1997. [AKW00] R.C. Alferness, H. Kogelnik, and T. H. Wood. The evolution of optical systems: Optics everywhere. Bell Labs Technical Journal, 5(1):188-202, Jan.-March 2000.

References

43

[Alf99] R. Alferness, editor. Bell Labs Technical Journal: Optical Networking, volume 4, Jan.-Mar. 1999. [Bak01] B. Bakhshi et al. 1 Tb/s (101 • 10 Gb/s) transmission over transpacific distance using 28 nm C-band EDFAs. In OFC 2001 Technical Digest, pages PD21/1-3, 2001. [Bar96] R.A. Barry, editor. IEEE Network: Special Issue on Optical Networks, volume 10, Nov. 1996. [Big01] S. Bigo et al. 10.2 Tb/s (256 x 42.7 Gbit/s PDM/WDM) transmission over 100 km TeraLight fiber with 1.28bit/s/Hz spectral efficiency. In OFC 2001 Technical Digest, pages PD25/1-3, 2001. [BKLW00] W. E Brinkman, T. L. Koch, D. V. Lang, and D. W. Wilt. The lasers behind the communications revolution. Bell Labs Technical Journal, 5(1 ):150-167, Jan.-March 2000. [Bra89] C.A. Brackett, editor. IEEE Communications Magazine: Special Issue on Lightwave Systems and Components, volume 27, Oct. 1989. [Cai01] J.-X. Cai et al. 2.4 Tb/s (120 x 20 Gb/s) transmission over transoceanic distance with optimum FEC overhead and 48% spectral efficiency. In OFC 2001 Technical Digest, pages PD20/1-3, 2001. [CHK+96] R.L. Cruz, G. R. Hill, A. L. Kellner, R. Ramaswami, and G. H. Sasaki, editors. IEEE JSAC/JLT Special Issue on Optical Networks, volume 14, June 1996. [CNW90] N.K. Cheung, G. Nosu, and G. Winzer, editors. IEEE JSAC: Special Issue on Dense WDM Networks, volume 8, Aug. 1990. [CS83] J.S. Cook and O. I. Szentisi. North American field trials and early applications in telephony. IEEE JSAC, 1:393-397, 1983. [CSH00] G.K. Chang, K. I. Sato, and D. K. Hunter, editors. 1EEEIOSA Journal of Lightwave Technology: Special Issue on Optical Networks, volume 18, 2000. [CT98] A.R. Chraplyvy and R. W. Tkach. Terabit/second transmission experiments. IEEE Journal of Quantum Electronics, 34(11):2103-2108, 1998. [DL00] S.S. Dixit and R J. Lin, editors. IEEE Communications Magazine: Optical Networks Come of Age, volume 38, Feb. 2000. [DYJ00] S.S. Dixit and A. Yla-Jaaski, editors. IEEE Communications Magazine: WDM Optical Networks: A Reality Check, volume 38, Mar. 2000. [FGO+96] M. Fujiwara, M. S. Goodman, M. J. O'Mahony, O. K. Tonguez, and A. E. Willner, editors. IEEE/OSA JLTIJSA C Special Issue on Multiwavelength Optical Technology and Networks, volume 14, June 1996.

44

INTRODUCTION TO OPTICAL NETWORKS

[Fra93] A.G. Fraser. Banquet speech. In Proceedings of Workshop on High-Performance Communication Subsystems, Williamsburg, VA, Sept. 1993. [Fuk01] K. Fukuchi et al. 10.92 Tb/s (273 x 40 Gb/s) triple-band/ultra-dense WDM optical-repeatered transmission experiment. In OFC 2001 Technical Digest, pages PD24/1-3, 2001. [GJR96] P.E. Green, E J. Janniello, and R. Ramaswami. Muitichannel protocol-transparent WDM distance extension using remodulation. IEEE JSA C/JLT Special Issue on Optical Networks, 14(6):962-967, June 1996. [Gla00] A.M. Glass et al. Advances in fiber optics. Bell Labs Technical Journal, 5(1):168-187, Jan.-March 2000. [GLM+00] O. Gerstel, B. Li, A. McGuire, G. Rouskas, K. Sivalingam, and Z. Zhang, editors. IEEE JSA C" Special Issue on Protocols and Architectures for Next-Generation Optical Networks, Oct. 2000. [Gna96] A.H. Gnauck et al. One terabit/s transmission experiment. In 0FC'96 Technical Digest, 1996. Postdeadline paper PD20. [Gre93] P.E. Green. Fiber-Optic Networks. Prentice Hall, Englewood Cliffs, NJ, 1993. [HD97] G.R. Hill and P. Demeester, editors. IEEE Communications Magazine: Special Issue on Photonic Networks in Europe, volume 35, April 1997. [Hec98] J. Hecht. Understanding Fiber Optics. Prentice Hall, Englewood Cliffs, NJ, 1998. [Hec99] J. Hecht. City of Light: The Story of Fiber Optics. Oxford University Press, New York, 1999. [HSS98] A.M. Hill, A. A. M. Saleh, and K. Sato, editors. IEEE JSAC" Special Issue on High-Capacity Optical Transport Networks, volume 16, Sept. 1998. [Ish83] H. Ishio. Japanese field trials and applications in telephony. IEEE JSAC, 1:404-412, 1983. [KH66] K.C. Kao and G. A. Hockham. Dielectric-fiber surface waveguides for optical frequencies. Proceedings of IEE, 133(3):1151-1158, July 1966. [KK97a] I.P. Kaminow and T. L. Koch, editors. Optical Fiber Telecommunications IIIA. Academic Press, San Diego, CA, 1997. [KK97b] I.P. Kaminow and T. L. Koch, editors. Optical Fiber Telecommunications IIIB. Academic Press, San Diego, CA, 1997. [KKM70] E P. Kapron, D. B. Keck, and R. D. Maurer. Radiation losses in glass optical waveguides. Applied Physics Letters, 17(10):423-425, Nov. 1970. [KLHN93] M.J. Karol, C. Lin, G. Hill, and K. Nosu, editors. IEEE/OSA Journal of Lightwave Technology: Special Issue on Broadband Optical Networks, May/June 1993.

References

45

[KM98] E W. Kerfoot and W. C. Marra. Undersea fiber optic networks: Past, present and future. IEEE JSA C" Special Issue on High-Capacity Optical Transport Networks, 16(7):1220-1225, Sept. 1998. [Kra99] J.M. Kraushaar. Fiber Deployment Update: End of Year 1998. Federal Communications Commission, Sept. 1999. Available from http://www.fcc.gov. [Lin89] C. Lin, editor. Optoelectronic Technology and Lightwave Communications Systems. Van Nostrand Reinhold, New York, 1989. [Mac74] J.B. MacChesney et al. Preparation of low-loss optical fibers using simultaneous vapor deposition and fusion. In Proceedings of l Oth International Congress on Glass, volume 6, pages 40-44, Kyoto, Japan, 1974. [MK88] S.D. Miller and I. P. Kaminow, editors. Optical Fiber Telecommunications II. Academic Press, San Diego, CA, 1988. [Mor96] T. Morioka et al. 100 Gb/s x 10 channel OTDM/WDM transmission using a single supercontinuum WDM source. In 0FC'96 Technical Digest, 1996. Postdeadline paper PD21. [MT83] A. Moncalvo and E Tosco. European field trials and early applications in telephony. IEEE JSAC, 1:398-403, 1983. [NO94] K. Nosu and M. J. O'Mahony, editors. IEEE Communications Magazine: Special Issue on Optically Multiplexed Networks, volume 32, Dec. 1994. [Ona96] H. Onaka et al. 1.1 Tb/s WDM transmission over a 150 km 1.3/~m zero-dispersion single-mode fiber. In 0FC'96 Technical Digest, 1996. Postdeadline paper PD19. [Pru89] P.R. Prucnal, editor. IEEE Network: Special Issue on Optical Multiaccess Networks, volume 3, March 1989. [RT84] P.K. Runge and P. R. Trischitta. The SL undersea lightwave system. IEEE/OSA Journal on Lightwave Technology, 2:744-753, 1984. [ST91] B.E.A. Saleh and M. C. Teich. Fundamentals of Photonics. Wiley, New York, 1991. [Sta83] J.R. Stauffer. FT3Cma lightwave system for metropolitan and intercity applications. IEEE JSAC, 1:413-419, 1983. [VPM01] G. Vareille, E Pitel, and J. E Marcerou. 3 Tb/s (300 • 11.6 Gbit/s) transmission over 7380 km using 28 nm C§ with 25 GHz channel spacing and NRZ format. In OFC 2001 Technical Digest, pages PD22/1-3, 2001. [Wil00] A.E. Willner, editor. IEEE Journal of Selected Topics in Quantum Electronics: Millennium Issue, volume 6, Nov./Dec. 2000.

46

INTRODUCTION TO OPTICAL NETWORKS

[Yan96] Y. Yano et al. 2.6 Tb/s WDM transmission experiment using optical duobinary coding. In Proceedings of European Conference on Optical Communication, 1996. Postdeadline paper Th.B.3.1. [Zhu01] B. Zhu et al. 3.08 Tb/s (77 x 42.7 Gb/s) transmission over 1200 km of non-zero dispersion-shifted fiber with 100-km spans using C- L-band distributed Raman amplification. In OFC 2001 Technical Digest, pages PD23/1-3, 2001.